Method for registering three-dimensional medical image and robot system

Abstract

The invention discloses a method for registering a three-dimensional medical image and a robot system, which is a measure for treating metal interference in a targeted manner and ensures that the position of a small ball can be accurately identified and registered even if non-target metal interference exists. The recognition precision of the registration ball is improved, the reliability and accuracy of the registration process are guaranteed, a robot system can complete coordinate system registration more rapidly and accurately before an operation, the method is crucial to a robot-assisted operation needing high-precision navigation, the working efficiency of the operation preparation stage is improved, and the operation cost is reduced. And meanwhile, the safety and the success rate of the operation can be improved.

Description

Technical Field

[0001] The present invention relates to the field of electromechanical equipment, and in particular to a method for registering three-dimensional medical images with a robot system. Background Art

[0002] In recent years, 3D C-arm systems have been increasingly used in surgical settings such as dentistry and orthopedics. Compared to traditional CT scanners, 3D C-arms are smaller and easier to transport, making them more suitable for surgical settings. Compared to traditional CT images, cone-beam computed tomography (CBCT) images acquired by 3D C-arms have higher spatial resolution, which improves navigation accuracy during robotic surgery. Furthermore, they have faster acquisition speeds, which reduces radiation dose to patients. Before robotic surgery can begin, registering the robot's system coordinates with the real-world coordinate system is essential. Coordinate registration typically involves matching landmarks on a registration plate. The landmarks are a set of small metal spheres embedded in the registration plate. During CBCT image acquisition, the registration plate is held by a robotic arm and simultaneously captured alongside the target surgical area. After identifying the distribution of the metal spheres in the CBCT image, a coordinate transformation is performed against a calibrated distribution template to generate the registration matrix.

[0003] Currently, metal sphere recognition mainly involves thresholding and connected domain screening. The thresholding method is a semi-automatic method that uses a fixed threshold for segmentation and then directly calculates the sphere center matching. If the matching is incorrect, the image is manually cropped to remove interference. This method is inefficient and has difficulty handling situations where interference overlaps with the metal sphere.

[0004] In summary, a method for registering three-dimensional medical images with robotic systems is needed to address the shortcomings of the existing technology. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the present invention provides a method for aligning three-dimensional medical images with a robotic system, aiming to solve the above problems.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for registering a three-dimensional medical image with a robotic system, comprising the following steps:

[0007] Step S1: Data preparation, the registration plate and the target surgical area are simultaneously acquired into the 3D image;

[0008] Step S2: read the 3D image, perform interpolation resampling, and read the pre-calibrated metal ball template center position and the actual ball diameter;

[0009] Step S3: Determine the target layer, obtain the initial screening threshold from the maximum value sorting of each slice layer by the maximum inter-class variance method, and determine the target layer for segmentation calculation;

[0010] Step S4: image segmentation threshold calculation, using the maximum inter-class variance method to calculate the segmentation threshold for the target layer, and segmenting the three-dimensional image according to the segmentation threshold;

[0011] Step S5: Image segmentation, normalizing the segmented three-dimensional image, setting an area threshold according to the maximum cross-sectional area, and obtaining connected domains smaller than the area threshold;

[0012] Step S6: clustering to calculate the sphere center, determine the number of clusters through clustering method, initialize the cluster center, and after clustering, select the class that meets the mentioned ratio and sphericity standards to obtain the metal ball recognition result;

[0013] Step S7: Calculate the distances between all the sphere centers actually identified, construct a feature vector, and match the feature vectors of the identified sphere center and the corresponding template sphere center;

[0014] Step S8: Using the singular value decomposition method to solve the transformation matrix between the two coordinate systems, and complete the coordinate system registration.

[0015] Optionally, the initial screening threshold is calculated in step S3 in the following manner:

[0016] Step A1: Calculate the maximum value of voxels in each slice layer of the 3D image, sort the maximum values of each layer, and form a maximum value list;

[0017] Step A2: Normalize the voxel values in the maximum value list to the range of [0, 255];

[0018] Step A3: Set t1 to any integer in the range [0, 255], count the number of elements N in the maximum value list that are less than or equal to t1, calculate the average value meanN of these N elements, count the number of elements M in the maximum value list that are greater than t1 and the average value meanM of these M elements;

[0019] Step A4: Traverse the range [0,255] and find t1 that maximizes Var to obtain the initial screening threshold.

[0020] Optionally, the target layer for segmentation calculation in step S3 is determined in the following manner:

[0021] Step B1: Set the layer with the maximum voxel value greater than the initial screening threshold as the initial screening layer, forming the initial screening layer group L, and count the number of layers N L , according to the number and diameter of metal balls, the theoretical distribution layer number Lt is obtained;

[0022] Step B2: Calculate the mean value meanL and standard deviation stdL of the maximum value list of the initial screening layer group L;

[0023] Step B3: Calculate the ratio of the mean meanL to the standard deviation and compare it with the metal threshold. If the ratio is greater than the metal threshold, the layer with the median of the maximum value of the voxels in the initial screening layer group L is taken as the target layer for subsequent calculation. Otherwise, the Xth layer from small to large is taken as the target layer for segmentation calculation.

[0024] Optionally, the segmentation threshold in step S4 is used to segment the three-dimensional image in the following manner:

[0025] The maximum inter-class variance method is used to calculate the segmentation threshold Ts for the target layer. The three-dimensional image is segmented as a whole using the segmentation threshold Ts. The voxel values less than Ts are set to 0, and the voxel values greater than or equal to Ts remain unchanged.

[0026] Optionally, the image segmentation in step S5 is performed in the following manner:

[0027] Step S51: normalizing the 3D image using the global maximum and minimum values of the 3D image;

[0028] Step S52: Calculate the theoretical maximum cross-sectional area Sa of the metal ball in the two-dimensional image based on the diameter of the metal ball and the physical pixel spacing of the image;

[0029] Step S53: setting an area threshold St, traversing each slice layer of the 3D image, calculating the 2D connected domains, retaining the connected domains with an area smaller than the area threshold, and obtaining a segmented 3D image.

[0030] Optionally, the cluster center is initialized in step S6 in the following manner:

[0031] Step S61: Calculate the distance threshold Td and randomly select the first cluster center C0;

[0032] Step S62: Calculate the distances from all remaining points to C0, and select the point with the largest distance to C0 as the second cluster center C1;

[0033] Step S63: Calculate the distances from the remaining points to the second cluster center, and select the point with the largest distance as the next cluster center Cx;

[0034] Step S64: Repeat step S63 until the distances from all points to the nearest cluster center are less than the distance threshold Td.

[0035] Optionally, the sphere center is calculated in the following manner:

[0036] Calculate the theoretical sphere volume in image space, perform volume screening and sphericity screening, sort various sphericities, select the class with the largest sphericity as the preset number of metal balls, obtain the metal ball recognition result, and use the center of each class as the recognition center.

[0037] Optionally, in step S7, the feature vectors of the identified sphere center and the corresponding template sphere center are matched in the following manner:

[0038] Step S71: Calculate the distance matrix based on the coordinates of each sphere center;

[0039] Step S72: Calculate the minimum value in the distance matrix except zero, which is the minimum sphere center distance;

[0040] Step S73: reorder the elements in the feature vector of the m-th ball from small to large;

[0041] Step S74: Calculate the distance matrix and each sphere eigenvector for both the identified sphere center and the template sphere center, calculate the distance between the two eigenvectors, and match the identified sphere center and the template sphere center corresponding to the two eigenvectors with the smallest distance to obtain a sphere center pair;

[0042] Step S75: Repeat step S74 to match the identified sphere center with the template sphere center one by one.

[0043] A three-dimensional medical image and robot system registration system adopts the three-dimensional medical image and robot system registration method, including a data acquisition module, a preprocessing module, a preliminary screening threshold calculation module, a target layer determination module, an image segmentation module, a normalization and connected domain extraction module, a cluster analysis module, a feature vector matching module, and a coordinate system registration module;

[0044] A data acquisition module is responsible for simultaneously acquiring three-dimensional image data of the registration plate and the target surgical area;

[0045] The preprocessing module is used to perform interpolation and resampling to make the image data suitable for further analysis;

[0046] The initial screening threshold calculation module is used to calculate the initial screening threshold from the maximum value sorting of each slice layer through the maximum inter-class variance method to determine which layers are the target layers for segmentation calculation;

[0047] The target layer determination module is used to determine the target layer that actually needs to be segmented based on the layer whose voxel maximum value is greater than the initial screening threshold.

[0048] Optionally, the image segmentation module is used to calculate a segmentation threshold for the target layer using the maximum inter-class variance method, and segment the three-dimensional image using the threshold;

[0049] The normalization and connected domain extraction module is used to normalize the segmented three-dimensional image and set an area threshold according to the maximum cross-sectional area to obtain connected domains with an area smaller than the threshold.

[0050] Cluster analysis module, which is used to determine the number of clusters and initialize the cluster centers through clustering methods, screen out the classes that meet the mentioned ratio and sphericity standards, and finally obtain the metal ball recognition results;

[0051] The feature vector matching module is used to calculate the distance between all the two sphere centers actually identified, construct the feature vector, and match the feature vector of the identified sphere center with the corresponding template sphere center;

[0052] The coordinate system registration module is used to solve the transformation matrix between the two coordinate systems using the singular value decomposition method to complete the coordinate system registration.

[0053] Beneficial effects of the present invention:

[0054] 1. In this invention, targeted measures to address metal interference ensure that the position of the registration ball can be accurately identified even in the presence of non-target metal interference. This not only improves the recognition accuracy of the registration ball, but also ensures the reliability and accuracy of the registration process;

[0055] 2. The present invention enables the robotic system to complete coordinate system registration more quickly and accurately before surgery, which is crucial for robot-assisted surgery that requires high-precision navigation. It improves the efficiency of the surgical preparation stage and also helps to improve the safety and success rate of the surgery.

[0056] 3. In the present invention, the maximum inter-class variance method is used to calculate the initial screening threshold and the segmentation threshold. This method can effectively separate the target metal balls from complex CBCT images, reducing the possibility of misjudgment. In addition, the threshold problem in three-dimensional space is converted through two two-dimensional threshold calculations, which reduces the computational complexity and eliminates the need for manual specification. This increases the degree of automation and adaptability of the method, making it more suitable for metal ball identification in different situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 The present invention is a flow chart of a method.

[0058] Figure 2 This is a schematic diagram of an original image primary screening layer of the present invention.

[0059] Figure 3 This is a schematic diagram of a binary image after image segmentation according to the present invention.

[0060] Figure 4 This is a schematic diagram of the results of a cluster analysis of the present invention.

[0061] Figure 5 This is a schematic diagram of the template sphere center matching result of the present invention.

[0062] Figure 6 This is a schematic diagram of a matching result of the present invention. DETAILED DESCRIPTION

[0063] In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0064] like Figures 1 to 6 As shown, a method for registering a three-dimensional medical image with a robotic system includes the following:

[0065] A registration plate was designed, inlaid with at least three small metal balls. The ball diameters and spacing were known, the spacing between each ball was unique, and the center position of each ball was precisely calibrated. Considering that metal artifacts in CBCT images can affect the accuracy of ball center detection, titanium balls were selected due to their low artifact yield. However, the response value of titanium balls is lower than that of common metals such as steel. Therefore, a more comprehensive metal interference detection mechanism is required in the presence of other metal interference in CBCT images.

[0066] Data Preparation

[0067] Acquire CBCT images, use a robotic arm to clamp the registration plate, and simultaneously capture all metal balls and the target surgical area into the CBCT image.

[0068] The CBCT 3D image is read and interpolated and resampled according to the physical pixel spacing of the 2D image so that the physical spacing of the 3D image slices is the same as the physical pixel spacing of the 2D image.

[0069] Read the center position and actual ball diameter of each metal ball template.

[0070] Threshold calculation

[0071] In order to reduce the amount of calculation, the threshold calculation in three-dimensional space is converted into two two-dimensional threshold calculations.

[0072] Target layer determination

[0073] Calculate the maximum value of voxels in each slice layer of the 3D image, sort the maximum values of each layer, and use the maximum inter-class variance method to calculate the initial screening threshold Tf. The specific calculation steps of the initial screening threshold Tf are as follows:

[0074] ① Normalize the maximum value list to the range of 0-255;

[0075] ② Let t1 be any integer in the range of 0-255, count the number of elements N in the maximum value list that are less than or equal to t1 and the average value meanN of these N elements, and count the number of elements M in the maximum value list that are greater than t1 and the average value meanM of these M elements;

[0076] ③Calculate the between-class variance, Var=(N(meanN-mean) 2 +M (meanM-mean) 2 ) / 256, mean is the mean of the maximum value list;

[0077] ④ Traverse the range of 0-255 and find t1 that makes var the largest, which is the initial screening threshold Tf.

[0078] The layers with the maximum voxel value greater than the initial screening threshold are retained as the initial screening layer group L, and the number of layers NL is counted. In addition, the theoretical distribution layer number Lt is calculated based on the number and diameter of metal balls. Due to the design of the registration plate and the shooting rules, no two target metal balls will appear in the same layer. The specific calculation steps for determining the target layer for segmentation calculation are as follows:

[0079] ① Calculate the mean value meanL and standard deviation stdL of the maximum value list of the initial screening layer group L;

[0080] ② Calculate the ratio of the mean value meanL to the standard deviation and compare it with the metal threshold. If the ratio is greater than the metal threshold, the layer with the median of the maximum value of the voxels in the initial screening layer group L is taken as the target layer for subsequent calculation. Otherwise, the Xth layer from small to large is taken as the target layer for segmentation calculation, where X = 0.6×Lt.

[0081] Calculate image segmentation threshold

[0082] The maximum inter-class variance method is used to calculate the segmentation threshold Ts for the target layer. The specific steps are the same as the maximum inter-class variance method mentioned above. The segmentation threshold Ts is applied to the entire 3D image. Voxels with values less than Ts are set to 0, and voxels with values greater than or equal to Ts are left unchanged.

[0083] Image Segmentation

[0084] Normalize the 3D image using the global maximum and minimum values of the 3D image.

[0085] The theoretical maximum cross-sectional area Sa of the metal ball in the two-dimensional image is calculated based on the diameter of the metal ball and the physical pixel spacing of the image. The area threshold St is set to 1.5Sa. Each slice layer of the three-dimensional image is traversed to calculate the two-dimensional connected domain. The connected domain with an area smaller than the area threshold St is retained to obtain the segmented three-dimensional image.

[0086] Kmeans clustering calculation of sphere center

[0087] Kmeans clustering is an unsupervised machine learning method. Its basic concept is to iteratively find a classification scheme with K (hyperparameter) clusters that minimizes the corresponding loss function. The loss function is generally defined as the sum of the squared errors between each sample and the center of the cluster to which it belongs. When the cluster centers are initialized optimally, Kmeans clustering is computationally much more efficient than three-dimensional connected domain extraction. Traditional Kmeans clustering requires input of the target number of clusters, K. This paper proposes an improved Kmeans clustering method that automatically determines the number of clusters, K. This is possible because, in the current problem, the target metal balls to be identified have high pixel density per ball and large distances between balls. Interfering metal objects also have similar characteristics.

[0088] The specific steps for cluster center initialization are as follows:

[0089] ① Calculate the distance threshold ThresDist = 3×(known sphere diameter D / spacing) and randomly select the first cluster center C0;

[0090] ② Calculate the distance from all remaining points to C0, and select the point with the largest distance from C0 as the second cluster center C1;

[0091] ③ Calculate the distances of all remaining points to the nearest cluster center, and the point with the largest distance is used as the next cluster center Cx;

[0092] ④ Repeat step ③ until the distance from all points to the nearest cluster center is less than the distance threshold ThresDist.

[0093] After initialization, the initial cluster centers are automatically determined based on the known sphere diameter. Using the number of initial cluster centers M as the hyperparameter K, Kmeans clustering is performed to obtain M well-divided clusters. Since distance judgment has already been performed during initialization, the clustering process converges quickly.

[0094] In the presence of interfering metal objects, there will be non-target metal ball classes in the clustering results, that is, M is greater than the preset number of metal balls N. At this time, it is necessary to further filter out the metal ball class. The specific steps of screening are as follows:

[0095] ① Calculate the theoretical sphere volume in image space V = 4π / 3*(0.5*D / spacing) 3 , D is the known ball diameter;

[0096] ② Volume screening: The number of points in a class is the class volume Vc. Calculate the class volume ratio Rv = Vc / V, and retain the classes with Rv in the interval (0.5, 1.3);

[0097] ③ Sphericity screening: Take the distance dc between the farthest point in the cluster and the cluster center, calculate the volume Vdc of the sphere with dc as the radius, let the sphericity of the cluster be Sc=Vc / Vdc, and retain the clusters with Sc greater than 0.5;

[0098] ④ Sort the various types of Sc and select the class with the largest number of Sc equal to the preset number of metal balls, which is the metal ball recognition result, and the center of each class is the recognized ball center.

[0099] Sphere center matching + calculation of coordinate system transformation

[0100] Calculate the distances between all the recognized sphere centers, record the minimum distance, and divide all the distances by the minimum distance. Each recognized sphere center now has a feature vector. Perform the same calculation on the center of the metal sphere template. The recognized sphere center and the template sphere center with matching feature vectors are matched. The specific calculation steps are as follows:

[0101] ① When the number of balls is n, the distance matrix is calculated based on the coordinates of the center of each ball.

[0102] , where d1n is the distance between the first ball and the nth ball (such as the Euclidean distance). It can be seen that the matrix is a symmetric matrix and all the values on the diagonal are 0;

[0103] ②Calculate the minimum value other than 0 in the distance matrix Adist, which is the minimum center distance distmin, and calculate ;

[0104] ② The mth ball (m≤n), its eigenvector is , reorder the elements in the eigenvector Vm so that they are arranged from small to large;

[0105] ③ Calculate the distance matrix and the eigenvectors of each sphere for both the identified sphere center and the template sphere center. Take the eigenvector Vrp of a certain identified sphere center and the eigenvector Vtq of a certain template sphere center, and calculate the Euclidean distance between the two eigenvectors. , the recognition sphere center and the template sphere center corresponding to the two eigenvectors with the smallest distance are the matching sphere center pair;

[0106] ⑤ Repeat step ④ to match the identified sphere center with the template sphere center one by one.

[0107] After the center of the identified sphere is correctly matched with the center of the template sphere, the singular value decomposition (SVD) method can be used to solve the transformation matrix RT between the two coordinate systems to complete the coordinate system registration.

[0108] The present invention incorporates targeted measures to address metal interference in each calculation step, ensuring accurate identification of the registration sphere's position even in the presence of non-target metal interference. This not only improves the accuracy of the registration sphere's recognition but also ensures the reliability and accuracy of the registration process. The Kmeans clustering method, a machine learning approach, is introduced to calculate the sphere's center. This method offers higher computational efficiency than traditional three-dimensional connected domain extraction. In particular, when the initial cluster centers are optimal, the clustering process converges quickly, significantly reducing computation time and improving the overall system's operational efficiency.

[0109] The ability to quickly and accurately complete coordinate system registration before surgery is crucial for robot-assisted surgery requiring high-precision navigation. This improves work efficiency during the surgical preparation phase, while also contributing to improved surgical safety and success rates. The maximum inter-class variance method is used to calculate the initial screening threshold and segmentation threshold. This method can effectively separate target metal balls from complex CBCT images, reducing the possibility of misjudgment. The method also transforms the threshold problem into a three-dimensional space through two two-dimensional threshold calculations, reducing computational complexity. The improved Kmeans clustering algorithm can automatically determine the optimal number of clusters, K, without manual specification. This increases the method's automation and adaptability, making it better suited for metal ball identification in different situations.

[0110] A three-dimensional medical image and robot system registration system adopts the three-dimensional medical image and robot system registration method, including a data acquisition module, a preprocessing module, a preliminary screening threshold calculation module, a target layer determination module, an image segmentation module, a normalization and connected domain extraction module, a cluster analysis module, a feature vector matching module, and a coordinate system registration module;

[0111] A data acquisition module is responsible for simultaneously acquiring three-dimensional image data of the registration plate and the target surgical area;

[0112] The preprocessing module is used to perform interpolation and resampling to make the image data suitable for further analysis;

[0113] The initial screening threshold calculation module is used to calculate the initial screening threshold from the maximum value sorting of each slice layer through the maximum inter-class variance method to determine which layers are the target layers for segmentation calculation;

[0114] The target layer determination module is used to determine the target layer that actually needs to be segmented based on the layer whose voxel maximum value is greater than the initial screening threshold.

[0115] Optionally, the image segmentation module is used to calculate a segmentation threshold for the target layer using the maximum inter-class variance method, and segment the three-dimensional image using the threshold;

[0116] The normalization and connected domain extraction module is used to normalize the segmented three-dimensional image and set an area threshold according to the maximum cross-sectional area to obtain connected domains with an area smaller than the threshold.

[0117] Cluster analysis module, which is used to determine the number of clusters and initialize the cluster centers through clustering methods, screen out the classes that meet the mentioned ratio and sphericity standards, and finally obtain the metal ball recognition results;

[0118] The feature vector matching module is used to calculate the distance between all the two sphere centers actually identified, construct the feature vector, and match the feature vector of the identified sphere center with the corresponding template sphere center;

[0119] The coordinate system registration module is used to solve the transformation matrix between the two coordinate systems using the singular value decomposition method to complete the coordinate system registration.

[0120] The embodiment described above is only a preferred solution of the present application and does not limit the present application in any form. There are other variations and modifications without exceeding the technical solution described in the claims.

Claims

1. A method for registering a three-dimensional medical image with a robotic system, characterized in that: The following steps are involved: Step S1: Data preparation, the registration plate and the target surgical area are simultaneously acquired into the 3D image; Step S2: read the 3D image, perform interpolation resampling, and read the pre-calibrated metal ball template center position and the actual ball diameter; Step S3: Determine the target layer, obtain the initial screening threshold from the maximum value sorting of each slice layer by the maximum inter-class variance method, and determine the target layer for segmentation calculation; Step S4: image segmentation threshold calculation, using the maximum inter-class variance method to calculate the segmentation threshold for the target layer, and segmenting the three-dimensional image according to the segmentation threshold; Step S5: Image segmentation, normalizing the segmented three-dimensional image, setting an area threshold according to the maximum cross-sectional area, and obtaining connected domains smaller than the area threshold; Step S6: clustering to calculate the sphere center, determine the number of clusters through clustering method, initialize the cluster center, and after clustering, select the class that meets the mentioned ratio and sphericity standards to obtain the metal ball recognition result; Step S7: Calculate the distances between all the sphere centers actually identified, construct a feature vector, and match the feature vectors of the identified sphere center and the corresponding template sphere center; Step S8: Using the singular value decomposition method to solve the transformation matrix between the two coordinate systems, and complete the coordinate system registration.

2. The method for registering a three-dimensional medical image with a robotic system according to claim 1, wherein: The initial screening threshold is calculated in step S3 in the following manner: Step A1: Calculate the maximum value of voxels in each slice layer of the 3D image, sort the maximum values of each layer, and form a maximum value list; Step A2: Normalize the voxel values in the maximum value list to the range of [0, 255]; Step A3: Set t1 to any integer in the range [0, 255], count the number of elements N in the maximum value list that are less than or equal to t1, calculate the average value meanN of these N elements, count the number of elements M in the maximum value list that are greater than t1 and the average value meanM of these M elements; Step A4: Traverse the range [0,255] and find t1 that maximizes Var to obtain the initial screening threshold.

3. The method for registering a three-dimensional medical image with a robotic system according to claim 1, wherein: In step S3, the target layer for segmentation calculation is determined by: Step B1: Set the layer with the maximum voxel value greater than the initial screening threshold as the initial screening layer, forming the initial screening layer group L, and count the number of layers N L , according to the number and diameter of metal balls, the theoretical distribution layer number Lt is obtained; Step B2: Calculate the mean value meanL and standard deviation stdL of the maximum value list of the initial screening layer group L; Step B3: Calculate the ratio of the mean meanL to the standard deviation and compare it with the metal threshold. If the ratio is greater than the metal threshold, the layer with the median of the maximum value of the voxels in the initial screening layer group L is taken as the target layer for subsequent calculation. Otherwise, the Xth layer from small to large is taken as the target layer for segmentation calculation.

4. The method for registering a three-dimensional medical image with a robotic system according to claim 1, wherein: The segmentation threshold in step S4 is used to segment the three-dimensional image in the following manner: The maximum inter-class variance method is used to calculate the segmentation threshold Ts for the target layer. The three-dimensional image is segmented as a whole using the segmentation threshold Ts. The voxel values less than Ts are set to 0, and the voxel values greater than or equal to Ts remain unchanged.

5. The method for registering a three-dimensional medical image with a robotic system according to claim 1, wherein: The image segmentation in step S5 is performed in the following manner: Step S51: normalizing the 3D image using the global maximum and minimum values of the 3D image; Step S52: Calculate the theoretical maximum cross-sectional area Sa of the metal ball in the two-dimensional image based on the diameter of the metal ball and the physical pixel spacing of the image; Step S53: setting an area threshold St, traversing each slice layer of the 3D image, calculating the 2D connected domains, retaining the connected domains with an area smaller than the area threshold, and obtaining a segmented 3D image.

6. The method for registering a three-dimensional medical image with a robotic system according to claim 1, wherein: In step S6, the cluster center is initialized in the following manner: Step S61: Calculate the distance threshold Td and randomly select the first cluster center C0; Step S62: Calculate the distances from all remaining points to C0, and select the point with the largest distance to C0 as the second cluster center C1; Step S63: Calculate the distances from the remaining points to the second cluster center, and select the point with the largest distance as the next cluster center Cx; Step S64: Repeat step S63 until the distances from all points to the nearest cluster center are less than the distance threshold Td.

7. The method for registering a three-dimensional medical image with a robotic system according to claim 6, wherein: The sphere center is calculated as follows: Calculate the theoretical sphere volume in image space, perform volume screening and sphericity screening, sort various sphericities, select the class with the largest sphericity as the preset number of metal balls, obtain the metal ball recognition result, and use the center of each class as the recognition center.

8. The method for registering a three-dimensional medical image with a robotic system according to claim 1, wherein: In step S7, the feature vectors of the identified sphere center and the corresponding template sphere center are matched in the following manner: Step S71: Calculate the distance matrix based on the coordinates of each sphere center; Step S72: Calculate the minimum value in the distance matrix except zero, which is the minimum sphere center distance; Step S73: reorder the elements in the feature vector of the m-th ball from small to large; Step S74: Calculate the distance matrix and each sphere eigenvector for both the identified sphere center and the template sphere center, calculate the distance between the two eigenvectors, and match the identified sphere center and the template sphere center corresponding to the two eigenvectors with the smallest distance to obtain a sphere center pair; Step S75: Repeat step S74 to match the identified sphere center with the template sphere center one by one.

9. A 3D medical image and robot system registration system, using the 3D medical image and robot system registration method according to claims 1-8, characterized in that: It includes data acquisition module, preprocessing module, initial screening threshold calculation module, target layer determination module, image segmentation module, normalization and connected domain extraction module, cluster analysis module, feature vector matching module, and coordinate system registration module; A data acquisition module is responsible for simultaneously acquiring three-dimensional image data of the registration plate and the target surgical area; The preprocessing module is used to perform interpolation and resampling to make the image data suitable for further analysis; The initial screening threshold calculation module is used to calculate the initial screening threshold from the maximum value sorting of each slice layer through the maximum inter-class variance method to determine which layers are the target layers for segmentation calculation; The target layer determination module is used to determine the target layer that actually needs to be segmented based on the layer whose voxel maximum value is greater than the initial screening threshold.

10. The method for registering a three-dimensional medical image with a robotic system according to claim 9, wherein: The image segmentation module is used to calculate the segmentation threshold of the target layer using the maximum inter-class variance method, and segment the three-dimensional image based on the threshold; The normalization and connected domain extraction module is used to normalize the segmented three-dimensional image and set an area threshold according to the maximum cross-sectional area to obtain connected domains with an area smaller than the threshold. Cluster analysis module, which is used to determine the number of clusters and initialize the cluster centers through clustering methods, screen out the classes that meet the mentioned ratio and sphericity standards, and finally obtain the metal ball recognition results; The feature vector matching module is used to calculate the distance between all the two sphere centers actually identified, construct the feature vector, and match the feature vector of the identified sphere center with the corresponding template sphere center; The coordinate system registration module is used to solve the transformation matrix between the two coordinate systems using the singular value decomposition method to complete the coordinate system registration.

Images